Stop asking AI questions. Start giving it work.

The tell

Open your AI history and read the last twenty things you sent. If most of them are questions, you are using a very expensive search engine.

I see this constantly. Someone says AI has not really changed how they work, I ask what they use it for, and the answer is a list of questions: what should I include in a proposal, how do I structure a performance review, what are good interview questions. All reasonable. All producing generic advice, because a question with no context can only be answered generically.

Then they conclude the tool is overrated. The tool is fine. It was asked for an opinion when it could have done the task.

Why asking feels safer

Three reasons, and they are all understandable.

It is what we are trained to do. Twenty-five years of search engines taught us that you put a question in a box and get information back. That muscle memory is strong and the interface looks similar.

Handing over work feels like it needs trust you have not built yet. Asking for advice is low stakes. Asking it to draft the thing you are going to send a client feels like a risk. So people stay in the shallow end, where the results are mild, which never builds the confidence to go deeper.

Giving work requires you to know what you want. This is the real barrier. "What should go in this proposal" is a question you can ask in four seconds. "Draft this proposal the way I would" requires you to have the call notes to hand, an example of your own work, and an opinion about what good looks like. That is five minutes of gathering, and we are all optimising for the four-second option.

Those five minutes are the whole game.

What a brief contains

When I hand work over rather than asking about it, the message has five parts. Not a formula to memorise, just what a competent colleague would need.

• The job. What should exist when this is done. A draft proposal. A summary a client can read. Six interview questions. Name the artifact.
• The materials. Your actual stuff. Past examples, the call notes, the export, the template. This is the part people skip and it is the part that matters most.
• The audience. Who reads this and what they already know. A proposal for a procurement team is a different document from one for a founder.
• The shape. How long, what format, what structure. "Same sections as the attached example" does an enormous amount of work in five words.
• The boundaries. What not to do. Do not invent pricing. Do not promise dates. Flag anything you are unsure of rather than guessing.

That last one deserves emphasis. Telling it what to do when it is uncertain is the single highest-value line in most briefs. Left alone, these systems fill gaps confidently. Told to flag gaps instead, they mostly will, and you get a draft with question marks in the right places rather than invented facts in the wrong ones.

Before and after

A proposal

Asking: "What should I include in a proposal for a corporate event?"

You get a list of standard sections you already knew.

Directing: "Attached are the last three proposals I sent and my notes from this morning's call. Draft the fourth in the same structure and voice. Use the pricing approach from the second one, it is closest in scale. Do not invent any numbers: where a price is needed and you cannot derive it from the examples, write TBC and tell me what you would need."

You get a draft you edit rather than a lecture about proposals.

A weekly report

Asking: "How should I analyse our sales data?"

Directing: "Here is this week's export and last month's summary that I wrote. Produce this week's in the same format. Call out anything that moved more than fifteen percent, and list the three things the team will most likely ask me about so I can prepare."

That last clause is the one people never think to add. You are not just asking for the report, you are asking it to anticipate the meeting.

Hiring

Asking: "What are good interview questions for an operations manager?"

Directing: "Here is the job description and short notes on our last two operations hires, one who worked out and one who did not. Write six questions that would have surfaced the difference between them. For each one, tell me what it is actually testing and what a weak answer sounds like."

The generic version gives you questions off the internet. This version gives you questions about your company.

Three habits that compound

Let it interview you first

When you are not sure how to brief something, say so and flip it: "I want to produce X. Before you start, ask me the five questions you most need answered to do it well."

This works unreasonably well. The questions it asks are usually the ones you had not thought about, and answering them produces the brief you could not write. It also surfaces, early, whether you actually have the material to do the job.

Stop starting from zero

Most people begin every session with an empty window. If you do the same kind of work repeatedly, keep the context somewhere it persists: a project, a saved document of your examples and standards, a file with your voice and your rules. Load it once, use it for months.

This is the single biggest gap between people who find AI mildly useful and people whose week actually changed. The second group is not writing better prompts. They stopped reintroducing themselves every morning.

Ask it to show its working

Add a line like "tell me which parts you were confident about and which you inferred." You get a map of where to focus your review, which turns checking from reading the whole thing suspiciously into reading two paragraphs carefully.

When a question is still the right move

None of this means questions are bad. They are right when:

• You genuinely do not know the landscape and need orientation before you can brief anything.
• You are pressure-testing your own thinking and want to be argued with.
• You want options rather than output, early in a decision.
• You are checking something factual and small.

The problem is not asking. The problem is only ever asking, and concluding from thin answers that the tool is thin.

Try it once this week

Take the task you are most dreading. Spend five minutes gathering the materials you would hand a new employee if they were doing it instead: an example of what good looks like, the raw inputs, and a sentence about who it is for. Hand all of it over with a clear statement of what should exist at the end, and a line telling it to flag anything it is unsure about rather than guessing.

What comes back will not be perfect. It will be a draft, and editing a draft is a different job from staring at an empty page. That is the whole difference, and it is worth considerably more than any prompt template.

If you want a hand structuring the brief, I built a free brief builder that walks you through the five parts and gives you something to paste into whichever AI you use. It runs entirely in your browser and nothing you type is sent anywhere.